Replace template with Audio2Tool leaderboard app

.gitignore

@@ -1,13 +0,0 @@
-auto_evals/
-venv/
-__pycache__/
-.env
-.ipynb_checkpoints
-*ipynb
-.vscode/
-
-eval-queue/
-eval-results/
-eval-queue-bk/
-eval-results-bk/
-logs/

.pre-commit-config.yaml

@@ -1,53 +0,0 @@
-# Copyright (c) 2022, NVIDIA CORPORATION.  All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-default_language_version:
-  python: python3
-
-ci:
-  autofix_prs: true
-  autoupdate_commit_msg: '[pre-commit.ci] pre-commit suggestions'
-  autoupdate_schedule: quarterly
-
-repos:
-  - repo: https://github.com/pre-commit/pre-commit-hooks
-    rev: v4.3.0
-    hooks:
-      - id: check-yaml
-      - id: check-case-conflict
-      - id: detect-private-key
-      - id: check-added-large-files
-        args: ['--maxkb=1000']
-      - id: requirements-txt-fixer
-      - id: end-of-file-fixer
-      - id: trailing-whitespace
-
-  - repo: https://github.com/PyCQA/isort
-    rev: 5.12.0
-    hooks:
-      - id: isort
-        name: Format imports
-
-  - repo: https://github.com/psf/black
-    rev: 22.12.0
-    hooks:
-      - id: black
-        name: Format code
-        additional_dependencies: ['click==8.0.2']
-
-  - repo: https://github.com/charliermarsh/ruff-pre-commit
-    # Ruff version.
-    rev: 'v0.0.267'
-    hooks:
-      - id: ruff

Makefile

@@ -1,13 +0,0 @@
-.PHONY: style format
-
-
-style:
-	python -m black --line-length 119 .
-	python -m isort .
-	ruff check --fix .
-
-
-quality:
-	python -m black --check --line-length 119 .
-	python -m isort --check-only .
-	ruff check .

README.md

@@ -1,48 +1,24 @@
 ---
-title: Audio2ToolLeaderboard
-emoji: 🥇
-colorFrom: green
-colorTo: indigo
+title: Audio2Tool Leaderboard
+emoji: 🎙️
+colorFrom: blue
+colorTo: green
 sdk: gradio
+sdk_version: 5.49.1
 app_file: app.py
 pinned: true
-license: cc-by-nc-4.0
-short_description: The 🤗 Audio2Tool Leaderboard evaluates cascaded & speechLLM
-sdk_version: 5.43.1
-tags:
-- leaderboard
+license: mit
+short_description: Voice-command tool calling across 8 difficulty tiers
 ---
 
-# Start the configuration
+# Audio2Tool Leaderboard
 
-Most of the variables to change for a default leaderboard are in `src/env.py` (replace the path for your leaderboard) and `src/about.py` (for tasks).
+Leaderboard for audio-language models on voice-command tool calling,
+evaluated on a held-out private subset of the
+[Audio2Tool dataset](https://huggingface.co/datasets/RVtech/Audio2Tool).
 
-Results files should have the following format and be stored as json files:
-```json
-{
-    "config": {
-        "model_dtype": "torch.float16", # or torch.bfloat16 or 8bit or 4bit
-        "model_name": "path of the model on the hub: org/model",
-        "model_sha": "revision on the hub",
-    },
-    "results": {
-        "task_name": {
-            "metric_name": score,
-        },
-        "task_name2": {
-            "metric_name": score,
-        }
-    }
-}
-```
+- Eval audio (no labels): [RVtech/Audio2Tool-private-eval](https://huggingface.co/datasets/RVtech/Audio2Tool-private-eval)
+- Benchmark code: [github.com/audio2tool/Audio2Tool](https://github.com/audio2tool/Audio2Tool)
 
-Request files are created automatically by this tool.
-
-If you encounter problem on the space, don't hesitate to restart it to remove the create eval-queue, eval-queue-bk, eval-results and eval-results-bk created folder.
-
-# Code logic for more complex edits
-
-You'll find 
-- the main table' columns names and properties in `src/display/utils.py`
-- the logic to read all results and request files, then convert them in dataframe lines, in `src/leaderboard/read_evals.py`, and `src/populate.py`
-- the logic to allow or filter submissions in `src/submission/submit.py` and `src/submission/check_validity.py`
+Submissions are scored inside this Space against hidden ground truth and
+stored in [RVtech/Audio2Tool-leaderboard-results](https://huggingface.co/datasets/RVtech/Audio2Tool-leaderboard-results).

app.py

@@ -1,204 +1,213 @@
+"""Audio2Tool Leaderboard Space."""
+
+import json
+import os
+import re
+import tempfile
+from datetime import datetime, timezone
+
 import gradio as gr
-from gradio_leaderboard import Leaderboard, ColumnFilter, SelectColumns
 import pandas as pd
-from apscheduler.schedulers.background import BackgroundScheduler
-from huggingface_hub import snapshot_download
-
-from src.about import (
-    CITATION_BUTTON_LABEL,
-    CITATION_BUTTON_TEXT,
-    EVALUATION_QUEUE_TEXT,
-    INTRODUCTION_TEXT,
-    LLM_BENCHMARKS_TEXT,
-    TITLE,
-)
-from src.display.css_html_js import custom_css
-from src.display.utils import (
-    BENCHMARK_COLS,
-    COLS,
-    EVAL_COLS,
-    EVAL_TYPES,
-    AutoEvalColumn,
-    ModelType,
-    fields,
-    WeightType,
-    Precision
-)
-from src.envs import API, EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH, QUEUE_REPO, REPO_ID, RESULTS_REPO, TOKEN
-from src.populate import get_evaluation_queue_df, get_leaderboard_df
-from src.submission.submit import add_new_eval
-
-
-def restart_space():
-    API.restart_space(repo_id=REPO_ID)
-
-### Space initialisation
-try:
-    print(EVAL_REQUESTS_PATH)
-    snapshot_download(
-        repo_id=QUEUE_REPO, local_dir=EVAL_REQUESTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
-    )
-except Exception:
-    restart_space()
-try:
-    print(EVAL_RESULTS_PATH)
-    snapshot_download(
-        repo_id=RESULTS_REPO, local_dir=EVAL_RESULTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
-    )
-except Exception:
-    restart_space()
-
-
-LEADERBOARD_DF = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS, BENCHMARK_COLS)
-
-(
-    finished_eval_queue_df,
-    running_eval_queue_df,
-    pending_eval_queue_df,
-) = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS)
-
-def init_leaderboard(dataframe):
-    if dataframe is None or dataframe.empty:
-        raise ValueError("Leaderboard DataFrame is empty or None.")
-    return Leaderboard(
-        value=dataframe,
-        datatype=[c.type for c in fields(AutoEvalColumn)],
-        select_columns=SelectColumns(
-            default_selection=[c.name for c in fields(AutoEvalColumn) if c.displayed_by_default],
-            cant_deselect=[c.name for c in fields(AutoEvalColumn) if c.never_hidden],
-            label="Select Columns to Display:",
-        ),
-        search_columns=[AutoEvalColumn.model.name, AutoEvalColumn.license.name],
-        hide_columns=[c.name for c in fields(AutoEvalColumn) if c.hidden],
-        filter_columns=[
-            ColumnFilter(AutoEvalColumn.model_type.name, type="checkboxgroup", label="Model types"),
-            ColumnFilter(AutoEvalColumn.precision.name, type="checkboxgroup", label="Precision"),
-            ColumnFilter(
-                AutoEvalColumn.params.name,
-                type="slider",
-                min=0.01,
-                max=150,
-                label="Select the number of parameters (B)",
-            ),
-            ColumnFilter(
-                AutoEvalColumn.still_on_hub.name, type="boolean", label="Deleted/incomplete", default=True
-            ),
-        ],
-        bool_checkboxgroup_label="Hide models",
-        interactive=False,
+from huggingface_hub import HfApi, hf_hub_download, snapshot_download
+
+from scoring import load_predictions_jsonl, score_predictions
+
+RESULTS_REPO = "RVtech/Audio2Tool-leaderboard-results"
+GT_REPO = "RVtech/Audio2Tool-leaderboard-gt"
+EVAL_REPO = "RVtech/Audio2Tool-private-eval"
+MAIN_DATASET = "RVtech/Audio2Tool"
+BENCHMARK_REPO = "https://github.com/audio2tool/Audio2Tool"
+
+TOKEN = os.environ.get("HF_TOKEN")
+api = HfApi(token=TOKEN)
+
+TIER_LABELS = {
+    "tier1_direct": "T1 Direct",
+    "tier2_parametric": "T2 Parametric",
+    "tier3_multi_intent": "T3 Multi-intent",
+    "tier4_implicit": "T4 Implicit",
+    "tier5_needle": "T5 Needle",
+    "tier6_correction": "T6 Correction",
+    "tier7_multiturn": "T7 Multi-turn",
+    "tier8_intent_blending": "T8 Blending",
+}
+
+
+def _load_gt():
+    path = hf_hub_download(GT_REPO, "gt.jsonl", repo_type="dataset", token=TOKEN)
+    with open(path) as f:
+        return [json.loads(line) for line in f if line.strip()]
+
+
+def _load_entries():
+    local = snapshot_download(RESULTS_REPO, repo_type="dataset",
+                              allow_patterns=["entries/*.json"], token=TOKEN)
+    entries = []
+    entries_dir = os.path.join(local, "entries")
+    if os.path.isdir(entries_dir):
+        for fname in os.listdir(entries_dir):
+            if fname.endswith(".json"):
+                with open(os.path.join(entries_dir, fname)) as f:
+                    entries.append(json.load(f))
+    # Keep the latest entry per model name
+    latest = {}
+    for e in entries:
+        key = e.get("model", "?")
+        if key not in latest or e.get("date", "") >= latest[key].get("date", ""):
+            latest[key] = e
+    return list(latest.values())
+
+
+def build_table():
+    rows = []
+    for e in _load_entries():
+        row = {
+            "Model": e.get("model", "?"),
+            "Type": e.get("model_type", ""),
+            "Tool Acc": e["overall"]["tool_accuracy"],
+            "Exact Match": e["overall"]["exact_match"],
+            "Param F1": e["overall"]["param_f1"],
+        }
+        for tier, label in TIER_LABELS.items():
+            t = e.get("tiers", {}).get(tier)
+            row[label] = t["tool_accuracy"] if t else None
+        row["Tiers"] = e["overall"].get("n_tiers", len(e.get("tiers", {})))
+        row["Date"] = e.get("date", "")
+        row["Submitter"] = e.get("submitter", "")
+        rows.append(row)
+    if not rows:
+        return pd.DataFrame(columns=["Model", "Type", "Tool Acc", "Exact Match", "Param F1"])
+    df = pd.DataFrame(rows).sort_values("Tool Acc", ascending=False).reset_index(drop=True)
+    df.insert(0, "Rank", range(1, len(df) + 1))
+    return df
+
+
+def submit(model_name, model_type, link, submitter, file):
+    try:
+        if not model_name or not model_name.strip():
+            return "**Error:** please provide a model name.", build_table()
+        if file is None:
+            return "**Error:** please upload a predictions JSONL file.", build_table()
+
+        predictions = load_predictions_jsonl(file)
+        gt_rows = _load_gt()
+        result = score_predictions(predictions, gt_rows)
+
+        entry = {
+            "model": model_name.strip(),
+            "model_type": model_type,
+            "submitter": (submitter or "").strip(),
+            "link": (link or "").strip(),
+            "date": datetime.now(timezone.utc).strftime("%Y-%m-%d"),
+            "seed_entry": False,
+            "tiers": result["tiers"],
+            "overall": result["overall"],
+        }
+
+        slug = re.sub(r"[^\w.-]+", "_", model_name.strip())
+        ts = datetime.now(timezone.utc).strftime("%Y%m%d_%H%M%S")
+        with tempfile.NamedTemporaryFile("w", suffix=".json", delete=False) as tmp:
+            json.dump(entry, tmp, indent=2)
+            tmp_path = tmp.name
+        api.upload_file(
+            path_or_fileobj=tmp_path,
+            path_in_repo=f"entries/{slug}_{ts}.json",
+            repo_id=RESULTS_REPO,
+            repo_type="dataset",
+            commit_message=f"Add submission: {model_name.strip()}",
+        )
+
+        o = result["overall"]
+        msg = (
+            f"**Submission scored and added.**\n\n"
+            f"- Tool Accuracy: **{o['tool_accuracy']:.1%}**\n"
+            f"- Exact Match: **{o['exact_match']:.1%}**\n"
+            f"- Param F1: **{o['param_f1']:.1%}**\n"
+            f"- Samples scored: {result['n_scored']} "
+            f"(missing predictions: {result['n_missing']}, tiers covered: {o['n_tiers']}/8)"
+        )
+        return msg, build_table()
+    except ValueError as e:
+        return f"**Error:** {e}", build_table()
+    except Exception as e:
+        return f"**Unexpected error:** {type(e).__name__}: {e}", build_table()
+
+
+ABOUT = f"""
+## Audio2Tool Leaderboard
+
+Benchmarks audio-language models on **voice-command tool calling**: the model
+hears an audio command and must emit the right tool call with the right
+parameters, across 8 difficulty tiers (direct commands, parametric, multi-intent,
+implicit, needle-in-haystack, self-correction, multi-turn, intent blending).
+
+- **Eval set:** [{EVAL_REPO}](https://huggingface.co/datasets/{EVAL_REPO}) —
+  1,579 held-out samples (~200 queries/tier) from the private split of
+  [{MAIN_DATASET}](https://huggingface.co/datasets/{MAIN_DATASET}).
+  Ground-truth labels are hidden; scoring happens in this Space.
+- **Benchmark code:** [{BENCHMARK_REPO}]({BENCHMARK_REPO})
+
+### Metrics
+
+| Metric | Definition |
+|--------|------------|
+| **Tool Acc** | Predicted primary tool name matches ground truth (case-insensitive) |
+| **Exact Match** | Tool name and all parameters match exactly |
+| **Param F1** | F1 over parameter key/value pairs (0 if the tool is wrong) |
+
+Overall scores are macro-averaged across covered tiers. Per-tier columns show
+Tool Accuracy. Entries marked *cascaded (oracle ASR)* consume ground-truth
+transcripts instead of audio and serve as text-side upper bounds.
+
+### How to submit
+
+1. Download the eval set: `hf download {EVAL_REPO} --repo-type dataset --local-dir eval`
+2. Run your model on each sample's audio (see `metadata.jsonl`; multi-turn
+   samples list turn audio in order).
+3. Produce a JSONL file, one row per sample:
+   `{{"sample_id": "tier1_direct/00042", "prediction": "setZoneTemperature(zone=Driver, temperature=21.0)"}}`
+   For multi-intent samples put several calls in one string (primary first).
+4. Upload it on the **Submit** tab.
+"""
+
+with gr.Blocks(title="Audio2Tool Leaderboard") as demo:
+    gr.Markdown("# 🎙️🔧 Audio2Tool Leaderboard")
+    gr.Markdown(
+        "Voice-command tool calling across 8 difficulty tiers — scored on a "
+        f"held-out private set ([eval audio]({'https://huggingface.co/datasets/' + EVAL_REPO}), "
+        f"[benchmark code]({BENCHMARK_REPO}))."
     )
 
-
-demo = gr.Blocks(css=custom_css)
-with demo:
-    gr.HTML(TITLE)
-    gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
-
-    with gr.Tabs(elem_classes="tab-buttons") as tabs:
-        with gr.TabItem("🏅 LLM Benchmark", elem_id="llm-benchmark-tab-table", id=0):
-            leaderboard = init_leaderboard(LEADERBOARD_DF)
-
-        with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=2):
-            gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")
-
-        with gr.TabItem("🚀 Submit here! ", elem_id="llm-benchmark-tab-table", id=3):
-            with gr.Column():
-                with gr.Row():
-                    gr.Markdown(EVALUATION_QUEUE_TEXT, elem_classes="markdown-text")
-
-                with gr.Column():
-                    with gr.Accordion(
-                        f"✅ Finished Evaluations ({len(finished_eval_queue_df)})",
-                        open=False,
-                    ):
-                        with gr.Row():
-                            finished_eval_table = gr.components.Dataframe(
-                                value=finished_eval_queue_df,
-                                headers=EVAL_COLS,
-                                datatype=EVAL_TYPES,
-                                row_count=5,
-                            )
-                    with gr.Accordion(
-                        f"🔄 Running Evaluation Queue ({len(running_eval_queue_df)})",
-                        open=False,
-                    ):
-                        with gr.Row():
-                            running_eval_table = gr.components.Dataframe(
-                                value=running_eval_queue_df,
-                                headers=EVAL_COLS,
-                                datatype=EVAL_TYPES,
-                                row_count=5,
-                            )
-
-                    with gr.Accordion(
-                        f"⏳ Pending Evaluation Queue ({len(pending_eval_queue_df)})",
-                        open=False,
-                    ):
-                        with gr.Row():
-                            pending_eval_table = gr.components.Dataframe(
-                                value=pending_eval_queue_df,
-                                headers=EVAL_COLS,
-                                datatype=EVAL_TYPES,
-                                row_count=5,
-                            )
-            with gr.Row():
-                gr.Markdown("# ✉️✨ Submit your model here!", elem_classes="markdown-text")
-
-            with gr.Row():
-                with gr.Column():
-                    model_name_textbox = gr.Textbox(label="Model name")
-                    revision_name_textbox = gr.Textbox(label="Revision commit", placeholder="main")
-                    model_type = gr.Dropdown(
-                        choices=[t.to_str(" : ") for t in ModelType if t != ModelType.Unknown],
-                        label="Model type",
-                        multiselect=False,
-                        value=None,
-                        interactive=True,
-                    )
-
-                with gr.Column():
-                    precision = gr.Dropdown(
-                        choices=[i.value.name for i in Precision if i != Precision.Unknown],
-                        label="Precision",
-                        multiselect=False,
-                        value="float16",
-                        interactive=True,
-                    )
-                    weight_type = gr.Dropdown(
-                        choices=[i.value.name for i in WeightType],
-                        label="Weights type",
-                        multiselect=False,
-                        value="Original",
-                        interactive=True,
-                    )
-                    base_model_name_textbox = gr.Textbox(label="Base model (for delta or adapter weights)")
-
-            submit_button = gr.Button("Submit Eval")
-            submission_result = gr.Markdown()
-            submit_button.click(
-                add_new_eval,
-                [
-                    model_name_textbox,
-                    base_model_name_textbox,
-                    revision_name_textbox,
-                    precision,
-                    weight_type,
-                    model_type,
-                ],
-                submission_result,
-            )
-
-    with gr.Row():
-        with gr.Accordion("📙 Citation", open=False):
-            citation_button = gr.Textbox(
-                value=CITATION_BUTTON_TEXT,
-                label=CITATION_BUTTON_LABEL,
-                lines=20,
-                elem_id="citation-button",
-                show_copy_button=True,
+    with gr.Tab("🏆 Leaderboard"):
+        table = gr.Dataframe(value=build_table, datatype="number", wrap=True)
+        refresh = gr.Button("Refresh")
+        refresh.click(fn=build_table, outputs=table)
+
+    with gr.Tab("📤 Submit"):
+        gr.Markdown(
+            "Score your model's predictions on the hidden test labels and add it "
+            "to the leaderboard. See the **About** tab for the prediction format."
+        )
+        with gr.Row():
+            model_name = gr.Textbox(label="Model name", placeholder="e.g. MyAudioLM-8B")
+            model_type = gr.Dropdown(
+                label="Model type",
+                choices=["end-to-end", "cascaded", "cascaded (oracle ASR)", "other"],
+                value="end-to-end",
             )
-
-scheduler = BackgroundScheduler()
-scheduler.add_job(restart_space, "interval", seconds=1800)
-scheduler.start()
-demo.queue(default_concurrency_limit=40).launch()
+        with gr.Row():
+            link = gr.Textbox(label="Link (paper / model card)", placeholder="https://…")
+            submitter = gr.Textbox(label="Submitted by", placeholder="your name or org")
+        file = gr.File(label="Predictions (.jsonl)", file_types=[".jsonl", ".json"], type="filepath")
+        submit_btn = gr.Button("Score and submit", variant="primary")
+        status = gr.Markdown()
+        submit_btn.click(
+            fn=submit,
+            inputs=[model_name, model_type, link, submitter, file],
+            outputs=[status, table],
+        )
+
+    with gr.Tab("ℹ️ About"):
+        gr.Markdown(ABOUT)
+
+demo.launch()

pyproject.toml

@@ -1,13 +0,0 @@
-[tool.ruff]
-# Enable pycodestyle (`E`) and Pyflakes (`F`) codes by default.
-select = ["E", "F"]
-ignore = ["E501"] # line too long (black is taking care of this)
-line-length = 119
-fixable = ["A", "B", "C", "D", "E", "F", "G", "I", "N", "Q", "S", "T", "W", "ANN", "ARG", "BLE", "COM", "DJ", "DTZ", "EM", "ERA", "EXE", "FBT", "ICN", "INP", "ISC", "NPY", "PD", "PGH", "PIE", "PL", "PT", "PTH", "PYI", "RET", "RSE", "RUF", "SIM", "SLF", "TCH", "TID", "TRY", "UP", "YTT"]
-
-[tool.isort]
-profile = "black"
-line_length = 119
-
-[tool.black]
-line-length = 119

requirements.txt

@@ -1,16 +1,3 @@
-APScheduler
-black
-datasets
-gradio
-gradio[oauth]
-gradio_leaderboard==0.0.13
-gradio_client
-huggingface-hub>=0.18.0
-matplotlib
-numpy
-pandas
-python-dateutil
-tqdm
-transformers
-tokenizers>=0.15.0
-sentencepiece
+gradio>=5.0
+huggingface_hub>=0.26
+pandas>=2.0

scoring.py

@@ -0,0 +1,166 @@
+"""
+Scoring logic for the Audio2Tool leaderboard.
+
+Mirrors the normalization and metrics of the benchmark code
+(github.com/audio2tool/Audio2Tool, audio_benchmark/evaluation/metrics.py)
+so leaderboard scores are directly comparable to local runs.
+"""
+
+import json
+import re
+from typing import Any, Dict, List, Tuple
+
+CALL_PATTERN = re.compile(r"(\w+)\s*\((.*?)\)")
+PARAM_PATTERN = re.compile(r"(\w+)\s*=\s*['\"]?([^,'\"]+)['\"]?")
+
+
+def parse_tool_calls(raw: str) -> List[Tuple[str, Dict[str, str]]]:
+    """Parse a prediction string into [(tool_name, params), ...].
+
+    Same regex as the benchmark's BaseAudioModel.parse_tool_call, extended
+    to capture multiple calls (for multi-intent tiers). The first call is
+    the primary tool.
+    """
+    calls = []
+    for m in CALL_PATTERN.finditer(raw or ""):
+        name = m.group(1)
+        params = {}
+        params_str = m.group(2)
+        if params_str.strip():
+            for pm in PARAM_PATTERN.finditer(params_str):
+                params[pm.group(1)] = pm.group(2).strip()
+        calls.append((name, params))
+    if not calls:
+        # Fallback: bare tool name(s) without parentheses
+        for word in (raw or "").split():
+            token = re.fullmatch(r"\w+", word.strip())
+            if token:
+                calls.append((token.group(0), {}))
+                break
+    return calls
+
+
+def normalize_tool_name(name: str) -> str:
+    if not name:
+        return ""
+    name = re.sub(r"\(.*\)", "", name)
+    return name.strip().lower()
+
+
+def normalize_parameters(params: Dict[str, Any]) -> Dict[str, str]:
+    if not params:
+        return {}
+    normalized = {}
+    for key, value in params.items():
+        norm_key = key.strip().lower()
+        if value is None:
+            norm_value = ""
+        elif isinstance(value, bool):
+            norm_value = "true" if value else "false"
+        elif isinstance(value, (int, float)):
+            norm_value = str(value)
+        else:
+            norm_value = str(value).strip().lower()
+        normalized[norm_key] = norm_value
+    return normalized
+
+
+def param_f1(predicted: Dict[str, Any], ground_truth: Dict[str, Any]) -> Tuple[float, bool]:
+    """Return (f1, exact_match) for parameter dicts."""
+    pred = normalize_parameters(predicted)
+    gt = normalize_parameters(ground_truth)
+
+    if not gt and not pred:
+        return 1.0, True
+    if not gt or not pred:
+        return 0.0, False
+
+    correct = sum(1 for k, v in gt.items() if pred.get(k) == v)
+    precision = correct / len(pred)
+    recall = correct / len(gt)
+    f1 = 2 * precision * recall / (precision + recall) if (precision + recall) else 0.0
+    exact = pred == gt
+    return f1, exact
+
+
+def score_predictions(
+    predictions: Dict[str, str],
+    gt_rows: List[Dict[str, Any]],
+) -> Dict[str, Any]:
+    """
+    Score {sample_id: prediction string} against ground-truth rows.
+
+    Returns per-tier and overall metrics plus coverage information.
+    """
+    by_tier: Dict[str, List[Dict[str, float]]] = {}
+    tier_totals: Dict[str, int] = {}
+    missing = []
+
+    for gt in gt_rows:
+        tier = gt["tier"]
+        tier_totals[tier] = tier_totals.get(tier, 0) + 1
+        raw = predictions.get(gt["sample_id"])
+        if raw is None:
+            missing.append(gt["sample_id"])
+            continue
+
+        calls = parse_tool_calls(raw)
+        pred_tool, pred_params = calls[0] if calls else ("", {})
+
+        tool_ok = normalize_tool_name(pred_tool) == normalize_tool_name(gt["tool_name"])
+        f1, exact_params = param_f1(pred_params, gt.get("extracted_params") or {})
+
+        by_tier.setdefault(tier, []).append({
+            "tool": float(tool_ok),
+            "exact": float(tool_ok and exact_params),
+            "f1": f1 if tool_ok else 0.0,
+        })
+
+    tiers = {}
+    for tier, scored in sorted(by_tier.items()):
+        n = len(scored)
+        tiers[tier] = {
+            "n": n,
+            "coverage": round(n / tier_totals[tier], 3),
+            "tool_accuracy": round(sum(s["tool"] for s in scored) / n, 4),
+            "exact_match": round(sum(s["exact"] for s in scored) / n, 4),
+            "param_f1": round(sum(s["f1"] for s in scored) / n, 4),
+        }
+
+    if not tiers:
+        raise ValueError("No predictions matched any eval sample_id.")
+
+    avg = lambda k: round(sum(t[k] for t in tiers.values()) / len(tiers), 4)
+    return {
+        "tiers": tiers,
+        "overall": {
+            "tool_accuracy": avg("tool_accuracy"),
+            "exact_match": avg("exact_match"),
+            "param_f1": avg("param_f1"),
+            "n_tiers": len(tiers),
+        },
+        "n_scored": sum(t["n"] for t in tiers.values()),
+        "n_missing": len(missing),
+    }
+
+
+def load_predictions_jsonl(path: str) -> Dict[str, str]:
+    """Load and validate a predictions JSONL file."""
+    predictions = {}
+    with open(path) as f:
+        for i, line in enumerate(f, 1):
+            line = line.strip()
+            if not line:
+                continue
+            try:
+                row = json.loads(line)
+            except json.JSONDecodeError as e:
+                raise ValueError(f"Line {i}: invalid JSON ({e})")
+            if "sample_id" not in row or "prediction" not in row:
+                raise ValueError(f"Line {i}: rows need 'sample_id' and 'prediction' keys")
+            if not isinstance(row["prediction"], str):
+                raise ValueError(f"Line {i}: 'prediction' must be a string")
+            predictions[row["sample_id"]] = row["prediction"]
+    if not predictions:
+        raise ValueError("File contains no predictions.")
+    return predictions

src/about.py

@@ -1,72 +0,0 @@
-from dataclasses import dataclass
-from enum import Enum
-
-@dataclass
-class Task:
-    benchmark: str
-    metric: str
-    col_name: str
-
-
-# Select your tasks here
-# ---------------------------------------------------
-class Tasks(Enum):
-    # task_key in the json file, metric_key in the json file, name to display in the leaderboard 
-    task0 = Task("anli_r1", "acc", "ANLI")
-    task1 = Task("logiqa", "acc_norm", "LogiQA")
-
-NUM_FEWSHOT = 0 # Change with your few shot
-# ---------------------------------------------------
-
-
-
-# Your leaderboard name
-TITLE = """<h1 align="center" id="space-title">Demo leaderboard</h1>"""
-
-# What does your leaderboard evaluate?
-INTRODUCTION_TEXT = """
-Intro text
-"""
-
-# Which evaluations are you running? how can people reproduce what you have?
-LLM_BENCHMARKS_TEXT = f"""
-## How it works
-
-## Reproducibility
-To reproduce our results, here is the commands you can run:
-
-"""
-
-EVALUATION_QUEUE_TEXT = """
-## Some good practices before submitting a model
-
-### 1) Make sure you can load your model and tokenizer using AutoClasses:
-```python
-from transformers import AutoConfig, AutoModel, AutoTokenizer
-config = AutoConfig.from_pretrained("your model name", revision=revision)
-model = AutoModel.from_pretrained("your model name", revision=revision)
-tokenizer = AutoTokenizer.from_pretrained("your model name", revision=revision)
-```
-If this step fails, follow the error messages to debug your model before submitting it. It's likely your model has been improperly uploaded.
-
-Note: make sure your model is public!
-Note: if your model needs `use_remote_code=True`, we do not support this option yet but we are working on adding it, stay posted!
-
-### 2) Convert your model weights to [safetensors](https://huggingface.co/docs/safetensors/index)
-It's a new format for storing weights which is safer and faster to load and use. It will also allow us to add the number of parameters of your model to the `Extended Viewer`!
-
-### 3) Make sure your model has an open license!
-This is a leaderboard for Open LLMs, and we'd love for as many people as possible to know they can use your model 🤗
-
-### 4) Fill up your model card
-When we add extra information about models to the leaderboard, it will be automatically taken from the model card
-
-## In case of model failure
-If your model is displayed in the `FAILED` category, its execution stopped.
-Make sure you have followed the above steps first.
-If everything is done, check you can launch the EleutherAIHarness on your model locally, using the above command without modifications (you can add `--limit` to limit the number of examples per task).
-"""
-
-CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results"
-CITATION_BUTTON_TEXT = r"""
-"""

src/display/css_html_js.py

@@ -1,105 +0,0 @@
-custom_css = """
-
-.markdown-text {
-    font-size: 16px !important;
-}
-
-#models-to-add-text {
-    font-size: 18px !important;
-}
-
-#citation-button span {
-    font-size: 16px !important;
-}
-
-#citation-button textarea {
-    font-size: 16px !important;
-}
-
-#citation-button > label > button {
-    margin: 6px;
-    transform: scale(1.3);
-}
-
-#leaderboard-table {
-    margin-top: 15px
-}
-
-#leaderboard-table-lite {
-    margin-top: 15px
-}
-
-#search-bar-table-box > div:first-child {
-    background: none;
-    border: none;
-}
- 
-#search-bar {
-    padding: 0px;
-}
-
-/* Limit the width of the first AutoEvalColumn so that names don't expand too much */
-#leaderboard-table td:nth-child(2),
-#leaderboard-table th:nth-child(2) {
-    max-width: 400px;
-    overflow: auto;
-    white-space: nowrap;
-}
-
-.tab-buttons button {
-    font-size: 20px;
-}
-
-#scale-logo {
-    border-style: none !important;
-    box-shadow: none;
-    display: block;
-    margin-left: auto;
-    margin-right: auto;
-    max-width: 600px;
-}
-
-#scale-logo .download {
-    display: none;
-}
-#filter_type{
-    border: 0;
-    padding-left: 0;
-    padding-top: 0;
-}
-#filter_type label {
-    display: flex;
-}
-#filter_type label > span{
-    margin-top: var(--spacing-lg);
-    margin-right: 0.5em;
-}
-#filter_type label > .wrap{
-    width: 103px;
-}
-#filter_type label > .wrap .wrap-inner{  
-    padding: 2px;
-}
-#filter_type label > .wrap .wrap-inner input{
-    width: 1px
-}
-#filter-columns-type{
-    border:0;
-    padding:0.5;
-}
-#filter-columns-size{
-    border:0;
-    padding:0.5;
-}
-#box-filter > .form{
-    border: 0
-}
-"""
-
-get_window_url_params = """
-    function(url_params) {
-        const params = new URLSearchParams(window.location.search);
-        url_params = Object.fromEntries(params);
-        return url_params;
-    }
-    """

src/display/formatting.py

@@ -1,27 +0,0 @@
-def model_hyperlink(link, model_name):
-    return f'<a target="_blank" href="{link}" style="color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;">{model_name}</a>'
-
-
-def make_clickable_model(model_name):
-    link = f"https://huggingface.co/{model_name}"
-    return model_hyperlink(link, model_name)
-
-
-def styled_error(error):
-    return f"<p style='color: red; font-size: 20px; text-align: center;'>{error}</p>"
-
-
-def styled_warning(warn):
-    return f"<p style='color: orange; font-size: 20px; text-align: center;'>{warn}</p>"
-
-
-def styled_message(message):
-    return f"<p style='color: green; font-size: 20px; text-align: center;'>{message}</p>"
-
-
-def has_no_nan_values(df, columns):
-    return df[columns].notna().all(axis=1)
-
-
-def has_nan_values(df, columns):
-    return df[columns].isna().any(axis=1)

src/display/utils.py

@@ -1,110 +0,0 @@
-from dataclasses import dataclass, make_dataclass
-from enum import Enum
-
-import pandas as pd
-
-from src.about import Tasks
-
-def fields(raw_class):
-    return [v for k, v in raw_class.__dict__.items() if k[:2] != "__" and k[-2:] != "__"]
-
-
-# These classes are for user facing column names,
-# to avoid having to change them all around the code
-# when a modif is needed
-@dataclass
-class ColumnContent:
-    name: str
-    type: str
-    displayed_by_default: bool
-    hidden: bool = False
-    never_hidden: bool = False
-
-## Leaderboard columns
-auto_eval_column_dict = []
-# Init
-auto_eval_column_dict.append(["model_type_symbol", ColumnContent, ColumnContent("T", "str", True, never_hidden=True)])
-auto_eval_column_dict.append(["model", ColumnContent, ColumnContent("Model", "markdown", True, never_hidden=True)])
-#Scores
-auto_eval_column_dict.append(["average", ColumnContent, ColumnContent("Average ⬆️", "number", True)])
-for task in Tasks:
-    auto_eval_column_dict.append([task.name, ColumnContent, ColumnContent(task.value.col_name, "number", True)])
-# Model information
-auto_eval_column_dict.append(["model_type", ColumnContent, ColumnContent("Type", "str", False)])
-auto_eval_column_dict.append(["architecture", ColumnContent, ColumnContent("Architecture", "str", False)])
-auto_eval_column_dict.append(["weight_type", ColumnContent, ColumnContent("Weight type", "str", False, True)])
-auto_eval_column_dict.append(["precision", ColumnContent, ColumnContent("Precision", "str", False)])
-auto_eval_column_dict.append(["license", ColumnContent, ColumnContent("Hub License", "str", False)])
-auto_eval_column_dict.append(["params", ColumnContent, ColumnContent("#Params (B)", "number", False)])
-auto_eval_column_dict.append(["likes", ColumnContent, ColumnContent("Hub ❤️", "number", False)])
-auto_eval_column_dict.append(["still_on_hub", ColumnContent, ColumnContent("Available on the hub", "bool", False)])
-auto_eval_column_dict.append(["revision", ColumnContent, ColumnContent("Model sha", "str", False, False)])
-
-# We use make dataclass to dynamically fill the scores from Tasks
-AutoEvalColumn = make_dataclass("AutoEvalColumn", auto_eval_column_dict, frozen=True)
-
-## For the queue columns in the submission tab
-@dataclass(frozen=True)
-class EvalQueueColumn:  # Queue column
-    model = ColumnContent("model", "markdown", True)
-    revision = ColumnContent("revision", "str", True)
-    private = ColumnContent("private", "bool", True)
-    precision = ColumnContent("precision", "str", True)
-    weight_type = ColumnContent("weight_type", "str", "Original")
-    status = ColumnContent("status", "str", True)
-
-## All the model information that we might need
-@dataclass
-class ModelDetails:
-    name: str
-    display_name: str = ""
-    symbol: str = "" # emoji
-
-
-class ModelType(Enum):
-    PT = ModelDetails(name="pretrained", symbol="🟢")
-    FT = ModelDetails(name="fine-tuned", symbol="🔶")
-    IFT = ModelDetails(name="instruction-tuned", symbol="⭕")
-    RL = ModelDetails(name="RL-tuned", symbol="🟦")
-    Unknown = ModelDetails(name="", symbol="?")
-
-    def to_str(self, separator=" "):
-        return f"{self.value.symbol}{separator}{self.value.name}"
-
-    @staticmethod
-    def from_str(type):
-        if "fine-tuned" in type or "🔶" in type:
-            return ModelType.FT
-        if "pretrained" in type or "🟢" in type:
-            return ModelType.PT
-        if "RL-tuned" in type or "🟦" in type:
-            return ModelType.RL
-        if "instruction-tuned" in type or "⭕" in type:
-            return ModelType.IFT
-        return ModelType.Unknown
-
-class WeightType(Enum):
-    Adapter = ModelDetails("Adapter")
-    Original = ModelDetails("Original")
-    Delta = ModelDetails("Delta")
-
-class Precision(Enum):
-    float16 = ModelDetails("float16")
-    bfloat16 = ModelDetails("bfloat16")
-    Unknown = ModelDetails("?")
-
-    def from_str(precision):
-        if precision in ["torch.float16", "float16"]:
-            return Precision.float16
-        if precision in ["torch.bfloat16", "bfloat16"]:
-            return Precision.bfloat16
-        return Precision.Unknown
-
-# Column selection
-COLS = [c.name for c in fields(AutoEvalColumn) if not c.hidden]
-
-EVAL_COLS = [c.name for c in fields(EvalQueueColumn)]
-EVAL_TYPES = [c.type for c in fields(EvalQueueColumn)]
-
-BENCHMARK_COLS = [t.value.col_name for t in Tasks]
-

src/envs.py

@@ -1,25 +0,0 @@
-import os
-
-from huggingface_hub import HfApi
-
-# Info to change for your repository
-# ----------------------------------
-TOKEN = os.environ.get("HF_TOKEN") # A read/write token for your org
-
-OWNER = "demo-leaderboard-backend" # Change to your org - don't forget to create a results and request dataset, with the correct format!
-# ----------------------------------
-
-REPO_ID = f"{OWNER}/leaderboard"
-QUEUE_REPO = f"{OWNER}/requests"
-RESULTS_REPO = f"{OWNER}/results"
-
-# If you setup a cache later, just change HF_HOME
-CACHE_PATH=os.getenv("HF_HOME", ".")
-
-# Local caches
-EVAL_REQUESTS_PATH = os.path.join(CACHE_PATH, "eval-queue")
-EVAL_RESULTS_PATH = os.path.join(CACHE_PATH, "eval-results")
-EVAL_REQUESTS_PATH_BACKEND = os.path.join(CACHE_PATH, "eval-queue-bk")
-EVAL_RESULTS_PATH_BACKEND = os.path.join(CACHE_PATH, "eval-results-bk")
-
-API = HfApi(token=TOKEN)

src/leaderboard/read_evals.py

@@ -1,196 +0,0 @@
-import glob
-import json
-import math
-import os
-from dataclasses import dataclass
-
-import dateutil
-import numpy as np
-
-from src.display.formatting import make_clickable_model
-from src.display.utils import AutoEvalColumn, ModelType, Tasks, Precision, WeightType
-from src.submission.check_validity import is_model_on_hub
-
-
-@dataclass
-class EvalResult:
-    """Represents one full evaluation. Built from a combination of the result and request file for a given run.
-    """
-    eval_name: str # org_model_precision (uid)
-    full_model: str # org/model (path on hub)
-    org: str 
-    model: str
-    revision: str # commit hash, "" if main
-    results: dict
-    precision: Precision = Precision.Unknown
-    model_type: ModelType = ModelType.Unknown # Pretrained, fine tuned, ...
-    weight_type: WeightType = WeightType.Original # Original or Adapter
-    architecture: str = "Unknown" 
-    license: str = "?"
-    likes: int = 0
-    num_params: int = 0
-    date: str = "" # submission date of request file
-    still_on_hub: bool = False
-
-    @classmethod
-    def init_from_json_file(self, json_filepath):
-        """Inits the result from the specific model result file"""
-        with open(json_filepath) as fp:
-            data = json.load(fp)
-
-        config = data.get("config")
-
-        # Precision
-        precision = Precision.from_str(config.get("model_dtype"))
-
-        # Get model and org
-        org_and_model = config.get("model_name", config.get("model_args", None))
-        org_and_model = org_and_model.split("/", 1)
-
-        if len(org_and_model) == 1:
-            org = None
-            model = org_and_model[0]
-            result_key = f"{model}_{precision.value.name}"
-        else:
-            org = org_and_model[0]
-            model = org_and_model[1]
-            result_key = f"{org}_{model}_{precision.value.name}"
-        full_model = "/".join(org_and_model)
-
-        still_on_hub, _, model_config = is_model_on_hub(
-            full_model, config.get("model_sha", "main"), trust_remote_code=True, test_tokenizer=False
-        )
-        architecture = "?"
-        if model_config is not None:
-            architectures = getattr(model_config, "architectures", None)
-            if architectures:
-                architecture = ";".join(architectures)
-
-        # Extract results available in this file (some results are split in several files)
-        results = {}
-        for task in Tasks:
-            task = task.value
-
-            # We average all scores of a given metric (not all metrics are present in all files)
-            accs = np.array([v.get(task.metric, None) for k, v in data["results"].items() if task.benchmark == k])
-            if accs.size == 0 or any([acc is None for acc in accs]):
-                continue
-
-            mean_acc = np.mean(accs) * 100.0
-            results[task.benchmark] = mean_acc
-
-        return self(
-            eval_name=result_key,
-            full_model=full_model,
-            org=org,
-            model=model,
-            results=results,
-            precision=precision,  
-            revision= config.get("model_sha", ""),
-            still_on_hub=still_on_hub,
-            architecture=architecture
-        )
-
-    def update_with_request_file(self, requests_path):
-        """Finds the relevant request file for the current model and updates info with it"""
-        request_file = get_request_file_for_model(requests_path, self.full_model, self.precision.value.name)
-
-        try:
-            with open(request_file, "r") as f:
-                request = json.load(f)
-            self.model_type = ModelType.from_str(request.get("model_type", ""))
-            self.weight_type = WeightType[request.get("weight_type", "Original")]
-            self.license = request.get("license", "?")
-            self.likes = request.get("likes", 0)
-            self.num_params = request.get("params", 0)
-            self.date = request.get("submitted_time", "")
-        except Exception:
-            print(f"Could not find request file for {self.org}/{self.model} with precision {self.precision.value.name}")
-
-    def to_dict(self):
-        """Converts the Eval Result to a dict compatible with our dataframe display"""
-        average = sum([v for v in self.results.values() if v is not None]) / len(Tasks)
-        data_dict = {
-            "eval_name": self.eval_name,  # not a column, just a save name,
-            AutoEvalColumn.precision.name: self.precision.value.name,
-            AutoEvalColumn.model_type.name: self.model_type.value.name,
-            AutoEvalColumn.model_type_symbol.name: self.model_type.value.symbol,
-            AutoEvalColumn.weight_type.name: self.weight_type.value.name,
-            AutoEvalColumn.architecture.name: self.architecture,
-            AutoEvalColumn.model.name: make_clickable_model(self.full_model),
-            AutoEvalColumn.revision.name: self.revision,
-            AutoEvalColumn.average.name: average,
-            AutoEvalColumn.license.name: self.license,
-            AutoEvalColumn.likes.name: self.likes,
-            AutoEvalColumn.params.name: self.num_params,
-            AutoEvalColumn.still_on_hub.name: self.still_on_hub,
-        }
-
-        for task in Tasks:
-            data_dict[task.value.col_name] = self.results[task.value.benchmark]
-
-        return data_dict
-
-
-def get_request_file_for_model(requests_path, model_name, precision):
-    """Selects the correct request file for a given model. Only keeps runs tagged as FINISHED"""
-    request_files = os.path.join(
-        requests_path,
-        f"{model_name}_eval_request_*.json",
-    )
-    request_files = glob.glob(request_files)
-
-    # Select correct request file (precision)
-    request_file = ""
-    request_files = sorted(request_files, reverse=True)
-    for tmp_request_file in request_files:
-        with open(tmp_request_file, "r") as f:
-            req_content = json.load(f)
-            if (
-                req_content["status"] in ["FINISHED"]
-                and req_content["precision"] == precision.split(".")[-1]
-            ):
-                request_file = tmp_request_file
-    return request_file
-
-
-def get_raw_eval_results(results_path: str, requests_path: str) -> list[EvalResult]:
-    """From the path of the results folder root, extract all needed info for results"""
-    model_result_filepaths = []
-
-    for root, _, files in os.walk(results_path):
-        # We should only have json files in model results
-        if len(files) == 0 or any([not f.endswith(".json") for f in files]):
-            continue
-
-        # Sort the files by date
-        try:
-            files.sort(key=lambda x: x.removesuffix(".json").removeprefix("results_")[:-7])
-        except dateutil.parser._parser.ParserError:
-            files = [files[-1]]
-
-        for file in files:
-            model_result_filepaths.append(os.path.join(root, file))
-
-    eval_results = {}
-    for model_result_filepath in model_result_filepaths:
-        # Creation of result
-        eval_result = EvalResult.init_from_json_file(model_result_filepath)
-        eval_result.update_with_request_file(requests_path)
-
-        # Store results of same eval together
-        eval_name = eval_result.eval_name
-        if eval_name in eval_results.keys():
-            eval_results[eval_name].results.update({k: v for k, v in eval_result.results.items() if v is not None})
-        else:
-            eval_results[eval_name] = eval_result
-
-    results = []
-    for v in eval_results.values():
-        try:
-            v.to_dict() # we test if the dict version is complete
-            results.append(v)
-        except KeyError:  # not all eval values present
-            continue
-
-    return results

src/populate.py

@@ -1,58 +0,0 @@
-import json
-import os
-
-import pandas as pd
-
-from src.display.formatting import has_no_nan_values, make_clickable_model
-from src.display.utils import AutoEvalColumn, EvalQueueColumn
-from src.leaderboard.read_evals import get_raw_eval_results
-
-
-def get_leaderboard_df(results_path: str, requests_path: str, cols: list, benchmark_cols: list) -> pd.DataFrame:
-    """Creates a dataframe from all the individual experiment results"""
-    raw_data = get_raw_eval_results(results_path, requests_path)
-    all_data_json = [v.to_dict() for v in raw_data]
-
-    df = pd.DataFrame.from_records(all_data_json)
-    df = df.sort_values(by=[AutoEvalColumn.average.name], ascending=False)
-    df = df[cols].round(decimals=2)
-
-    # filter out if any of the benchmarks have not been produced
-    df = df[has_no_nan_values(df, benchmark_cols)]
-    return df
-
-
-def get_evaluation_queue_df(save_path: str, cols: list) -> list[pd.DataFrame]:
-    """Creates the different dataframes for the evaluation queues requestes"""
-    entries = [entry for entry in os.listdir(save_path) if not entry.startswith(".")]
-    all_evals = []
-
-    for entry in entries:
-        if ".json" in entry:
-            file_path = os.path.join(save_path, entry)
-            with open(file_path) as fp:
-                data = json.load(fp)
-
-            data[EvalQueueColumn.model.name] = make_clickable_model(data["model"])
-            data[EvalQueueColumn.revision.name] = data.get("revision", "main")
-
-            all_evals.append(data)
-        elif ".md" not in entry:
-            # this is a folder
-            sub_entries = [e for e in os.listdir(f"{save_path}/{entry}") if os.path.isfile(e) and not e.startswith(".")]
-            for sub_entry in sub_entries:
-                file_path = os.path.join(save_path, entry, sub_entry)
-                with open(file_path) as fp:
-                    data = json.load(fp)
-
-                data[EvalQueueColumn.model.name] = make_clickable_model(data["model"])
-                data[EvalQueueColumn.revision.name] = data.get("revision", "main")
-                all_evals.append(data)
-
-    pending_list = [e for e in all_evals if e["status"] in ["PENDING", "RERUN"]]
-    running_list = [e for e in all_evals if e["status"] == "RUNNING"]
-    finished_list = [e for e in all_evals if e["status"].startswith("FINISHED") or e["status"] == "PENDING_NEW_EVAL"]
-    df_pending = pd.DataFrame.from_records(pending_list, columns=cols)
-    df_running = pd.DataFrame.from_records(running_list, columns=cols)
-    df_finished = pd.DataFrame.from_records(finished_list, columns=cols)
-    return df_finished[cols], df_running[cols], df_pending[cols]

src/submission/check_validity.py

@@ -1,99 +0,0 @@
-import json
-import os
-import re
-from collections import defaultdict
-from datetime import datetime, timedelta, timezone
-
-import huggingface_hub
-from huggingface_hub import ModelCard
-from huggingface_hub.hf_api import ModelInfo
-from transformers import AutoConfig
-from transformers.models.auto.tokenization_auto import AutoTokenizer
-
-def check_model_card(repo_id: str) -> tuple[bool, str]:
-    """Checks if the model card and license exist and have been filled"""
-    try:
-        card = ModelCard.load(repo_id)
-    except huggingface_hub.utils.EntryNotFoundError:
-        return False, "Please add a model card to your model to explain how you trained/fine-tuned it."
-
-    # Enforce license metadata
-    if card.data.license is None:
-        if not ("license_name" in card.data and "license_link" in card.data):
-            return False, (
-                "License not found. Please add a license to your model card using the `license` metadata or a"
-                " `license_name`/`license_link` pair."
-            )
-
-    # Enforce card content
-    if len(card.text) < 200:
-        return False, "Please add a description to your model card, it is too short."
-
-    return True, ""
-
-def is_model_on_hub(model_name: str, revision: str, token: str = None, trust_remote_code=False, test_tokenizer=False) -> tuple[bool, str]:
-    """Checks if the model model_name is on the hub, and whether it (and its tokenizer) can be loaded with AutoClasses."""
-    try:
-        config = AutoConfig.from_pretrained(model_name, revision=revision, trust_remote_code=trust_remote_code, token=token)
-        if test_tokenizer:
-            try:
-                tk = AutoTokenizer.from_pretrained(model_name, revision=revision, trust_remote_code=trust_remote_code, token=token)
-            except ValueError as e:
-                return (
-                    False,
-                    f"uses a tokenizer which is not in a transformers release: {e}",
-                    None
-                )
-            except Exception as e:
-                return (False, "'s tokenizer cannot be loaded. Is your tokenizer class in a stable transformers release, and correctly configured?", None)
-        return True, None, config
-
-    except ValueError:
-        return (
-            False,
-            "needs to be launched with `trust_remote_code=True`. For safety reason, we do not allow these models to be automatically submitted to the leaderboard.",
-            None
-        )
-
-    except Exception as e:
-        return False, "was not found on hub!", None
-
-
-def get_model_size(model_info: ModelInfo, precision: str):
-    """Gets the model size from the configuration, or the model name if the configuration does not contain the information."""
-    try:
-        model_size = round(model_info.safetensors["total"] / 1e9, 3)
-    except (AttributeError, TypeError):
-        return 0  # Unknown model sizes are indicated as 0, see NUMERIC_INTERVALS in app.py
-
-    size_factor = 8 if (precision == "GPTQ" or "gptq" in model_info.modelId.lower()) else 1
-    model_size = size_factor * model_size
-    return model_size
-
-def get_model_arch(model_info: ModelInfo):
-    """Gets the model architecture from the configuration"""
-    return model_info.config.get("architectures", "Unknown")
-
-def already_submitted_models(requested_models_dir: str) -> set[str]:
-    """Gather a list of already submitted models to avoid duplicates"""
-    depth = 1
-    file_names = []
-    users_to_submission_dates = defaultdict(list)
-
-    for root, _, files in os.walk(requested_models_dir):
-        current_depth = root.count(os.sep) - requested_models_dir.count(os.sep)
-        if current_depth == depth:
-            for file in files:
-                if not file.endswith(".json"):
-                    continue
-                with open(os.path.join(root, file), "r") as f:
-                    info = json.load(f)
-                    file_names.append(f"{info['model']}_{info['revision']}_{info['precision']}")
-
-                    # Select organisation
-                    if info["model"].count("/") == 0 or "submitted_time" not in info:
-                        continue
-                    organisation, _ = info["model"].split("/")
-                    users_to_submission_dates[organisation].append(info["submitted_time"])
-
-    return set(file_names), users_to_submission_dates

src/submission/submit.py

@@ -1,119 +0,0 @@
-import json
-import os
-from datetime import datetime, timezone
-
-from src.display.formatting import styled_error, styled_message, styled_warning
-from src.envs import API, EVAL_REQUESTS_PATH, TOKEN, QUEUE_REPO
-from src.submission.check_validity import (
-    already_submitted_models,
-    check_model_card,
-    get_model_size,
-    is_model_on_hub,
-)
-
-REQUESTED_MODELS = None
-USERS_TO_SUBMISSION_DATES = None
-
-def add_new_eval(
-    model: str,
-    base_model: str,
-    revision: str,
-    precision: str,
-    weight_type: str,
-    model_type: str,
-):
-    global REQUESTED_MODELS
-    global USERS_TO_SUBMISSION_DATES
-    if not REQUESTED_MODELS:
-        REQUESTED_MODELS, USERS_TO_SUBMISSION_DATES = already_submitted_models(EVAL_REQUESTS_PATH)
-
-    user_name = ""
-    model_path = model
-    if "/" in model:
-        user_name = model.split("/")[0]
-        model_path = model.split("/")[1]
-
-    precision = precision.split(" ")[0]
-    current_time = datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")
-
-    if model_type is None or model_type == "":
-        return styled_error("Please select a model type.")
-
-    # Does the model actually exist?
-    if revision == "":
-        revision = "main"
-
-    # Is the model on the hub?
-    if weight_type in ["Delta", "Adapter"]:
-        base_model_on_hub, error, _ = is_model_on_hub(model_name=base_model, revision=revision, token=TOKEN, test_tokenizer=True)
-        if not base_model_on_hub:
-            return styled_error(f'Base model "{base_model}" {error}')
-
-    if not weight_type == "Adapter":
-        model_on_hub, error, _ = is_model_on_hub(model_name=model, revision=revision, token=TOKEN, test_tokenizer=True)
-        if not model_on_hub:
-            return styled_error(f'Model "{model}" {error}')
-
-    # Is the model info correctly filled?
-    try:
-        model_info = API.model_info(repo_id=model, revision=revision)
-    except Exception:
-        return styled_error("Could not get your model information. Please fill it up properly.")
-
-    model_size = get_model_size(model_info=model_info, precision=precision)
-
-    # Were the model card and license filled?
-    try:
-        license = model_info.cardData["license"]
-    except Exception:
-        return styled_error("Please select a license for your model")
-
-    modelcard_OK, error_msg = check_model_card(model)
-    if not modelcard_OK:
-        return styled_error(error_msg)
-
-    # Seems good, creating the eval
-    print("Adding new eval")
-
-    eval_entry = {
-        "model": model,
-        "base_model": base_model,
-        "revision": revision,
-        "precision": precision,
-        "weight_type": weight_type,
-        "status": "PENDING",
-        "submitted_time": current_time,
-        "model_type": model_type,
-        "likes": model_info.likes,
-        "params": model_size,
-        "license": license,
-        "private": False,
-    }
-
-    # Check for duplicate submission
-    if f"{model}_{revision}_{precision}" in REQUESTED_MODELS:
-        return styled_warning("This model has been already submitted.")
-
-    print("Creating eval file")
-    OUT_DIR = f"{EVAL_REQUESTS_PATH}/{user_name}"
-    os.makedirs(OUT_DIR, exist_ok=True)
-    out_path = f"{OUT_DIR}/{model_path}_eval_request_False_{precision}_{weight_type}.json"
-
-    with open(out_path, "w") as f:
-        f.write(json.dumps(eval_entry))
-
-    print("Uploading eval file")
-    API.upload_file(
-        path_or_fileobj=out_path,
-        path_in_repo=out_path.split("eval-queue/")[1],
-        repo_id=QUEUE_REPO,
-        repo_type="dataset",
-        commit_message=f"Add {model} to eval queue",
-    )
-
-    # Remove the local file
-    os.remove(out_path)
-
-    return styled_message(
-        "Your request has been submitted to the evaluation queue!\nPlease wait for up to an hour for the model to show in the PENDING list."
-    )